Perception & Psychophysics
1998.60 (8). /384-1405
Multiple-feature discrimination faster than singlefeature discrimination within the same object?
USA R. FOURNIER
Washington State University, PuUman, Washington
CHARLES W. ERIKSEN
University ofIUinois at Urbana-Champaign, Champaign, IUinois
and
CHRISTOPHER BOWD
Washington State University, PuUman, Washington
In the present study, we investigated whether judging the presence of multiple features within an object would be superior to judging the presence of only one feature. Feature discriminability and the
number of features to discriminate within an object were varied. Specific features were judged as present or absent. Results showed thatjudging the presence oftwo or three features was faster thanjudging the presence of the less discriminable of these two or three features alone (multiple-feature benefits). These findings suggest that relevant features within an object activate (prime) adecision or
response in a parallel, asynchronous fashion based on discriminability (Miller, 1982a). The ability of a
response priming model, a response mapping model, and a template model to account for multiplefeature benefits is discussed.
It is generally assumed that task complexity increases
when two or more features, relative to one feature, must
be determined to match a particular stimulus category
(Wickens, 1984). Such multifeature judgments are typical in tasks of pattern recognition. Common sense suggests that comparing more information may require more
time or may be time limited by the most difficult comparison. Such predictions are made by simple serial and
parallel processing models, respectively, that assume that
features within an object are processed in an independent, discrete fashion (e.g., Nickerson, 1972; Sternberg,
1969). However, there is evidence that is inconsistent
with such predictions. For example, in same-different
tasks, it is typically found that same judgments are faster
than different judgments, even though same judgments
require that all object features be judged as same, whereas
different judgments require that only one feature be
judged as different (e.g., Bamber, 1969; Hawkins, 1969;
Miller & Bauer, 1981; Nickerson, 1972; Proctor, 1981).
In the present study, we investigated another empirical
phenomenon that also appears inconsistent with simple
serial and parallel models.
Experiments land 2 of this research were supported by Public
Health Service Research Career Program Award K6-MH-22014 to the
second author and by United States Public Health Service Research
Grant MH-OI 206 also to the second author. The authors thank Gordon
Logan, Tom Carr, Bill Bacon, and Toby Mordkoff for their invaluable
comments conceming this manuscript. Correspondence should be addressed to L. R. Fournier, Department ofPsychology, Washington State
University, Pullman, WA 99164-4820 (e-mail: fournier@wsunix.
wsu.edu).
Copyright 1998 Psychonomic Society, Inc.
In preliminary experiments, we have found that judging the presence oftwo or more features within an object
can be faster than judging the presence of a single feature
alone. We first encountered this phenomenon when attempting to vary attention processing demands by varying both feature discrimination and the number of target
features to attend to within a single object (see Fournier
& C. W. Eriksen, 1991). The object was a letter stimulus
that varied across the three attributes of color (red or
green), shape (the letter S or the letter C), and size (large
or small). Discrimination difficulty varied across these
three different features: The color discrimination was relatively easy, the shape discrimination was more difficult
(because the letters shared similar features), and the size
discrimination was the most difficult (because the large
stimulus was only 22% larger than the small stimulus).
The number oftarget features (one, two, or three) within
the object also varied. The observers judged the presence
or absence of one or more features of an object letter embedded in a circular array of distractor letters. The location ofthe relevant object letter was cued by a bar marker
either simultaneously with or at various intervals before
letter display onset.
Consistent with expectations, we found that "present"
and "absent" responses among the one- feature j udgments
were fastest and most accurate for color, intermediate in
speed and accuracy for shape, and slowest and least accurate for size. But, surprisingly, we found that judging
the presence of combined size and color (two features)
and combined shape and size (two features) was faster
and more accurate thanjudging the presence of size alone
1384
FEATURE DISCRIMINATION
1385
(one feature). In addition, judging the presence of com- made that indicated whether the target features were prebined shape and color (two features) was faster and more sent or absent.
accurate than judging the presence of shape alone (one
feature). Moreover, judging the presence of all three fea- Method
Subjects. Four male and 4 female undergraduates at the Univertures ofsize, color, and shape was faster thanjudging the
presence of size alone, shape alone, and combined size sity ofiliinois served as paid volunteers. All were right-handed, and
all reported having normal or corrected-to-normal vision.
and shape (Fournier & C. W. Eriksen, 1991).
Apparatus and Stimuli. Stimuli were presented on a computer
These findings suggest that judging the presence of monitor. In order to keep stimulus distance constant across subjects,
two or more target features within an object can be faster the subjects viewed the stimuli through a face mask. A white fixaand more accurate thanjudging the presence ofthe least tion cross measuring 0.20° ofvisual angle appeared in the center of
discriminable of these target features alone. This phe- the monitor before and during each trial.
The letter probe appeared 0.37° ofvisual angle above the fixation
nomen on will be referred to as multiple-feature benefits.
It is unclear as to how processing more information led cross. The features of the letter probe varied on three different dimensions: shape (S or C), size (large or smalI), and color (red or
to faster and more accurate responses. In order to make green). The large letters and the small letters were, respectively,
a "present" response in the three- feature judgments, one 0.33° and 0.27° of visual angle in height (a 22% size difference).
must determine whether a specific size, a specific shape, Luminance for green-colored letters was 13.7 cd/m-; luminance for
and a specific color are present. However, in order to make red-colored letters was 9.6 cd/m-.
A written message indicating the target features to be judged apa "present" response in the one-feature judgments ofsize
alone or shape alone, only one target feature must be peared in the upper left half of the monitor. When letter shape was
a target, the actualletter character appeared in the message, because
determined.
it could not be spelled out. In addition, when letter shape was a tarThese findings are inconsistent with simple serial or get and the size of the letter was not, both the small and the large
parallel feature processing models (see also findings by version of the letter character appeared in the message along with
Lockhead, 1972, 1979, and Miller, 1978). Simple serial the word or.
The subject initiated each trial by pressing a hand button held in
models predict that increasing the number of target features within an object should increase "present" response the left hand. He/she moved a hand lever with the right hand to the
RTs. In other words, judging whether three specific fea- left or the right to indicate whether the specific target feature( s) was
present or absent. The computer recorded response time (RT) and
tures are present should take longer than judging whether response accuracy.
only a single feature is present. This is because all three
Procedure. At the start of each trial, a message was displayed infeatures must be determined, as opposed to only one fea- dicating which specific one, two, or three target features the subject
ture, before a "present" response can be correctly executed was to judge as present or absent. This message was displayed until
(without guessing). Simple parallel models, in contrast, the subject pressed the hand button to clear the screen. After the
predict thatjudging the presence ofthree features should screen was cleared, the subject saw a small circle 0.20° of visual
angle in the center of the monitor. Approximately I sec later, this
take as long as judging the presence ofthe slowest (least circle was replaced by a fixation cross. The subject was to fixate the
discriminable) of any one of these three features. Thus, cross before initiating a trial. The subject initiated each trial by
"present" response latency would be limited by the target pressing the hand button when the fixation cross was in clear focus.
Immediately after pressing the hand button, the letter probe apfeature that has the slowest processing time.
In the present study, we examined the phenomenon of peared above the fixation cross for a duration of 50 msec.
The subjects judged whether the target features were present or
multiple-feature benefits. We chose to evaluate feature
absent. A "present" response was required when all of the target
processing within a single object because we could keep the features were present. An "absent" response was required when any
visual stimulus constant and vary only the features within one of'the target features was absent. Half ofthe subjects moved the
this stimulus that were relevant for a particular response. hand lever to the right when the target feature( s) was present and to
The ability ofvarious information processing models to the left when the target feature(s) was absent. The other half ofthe
subjects had the opposite hand-lever response assignment. The subaccount for multiple-feature benefits is discussed.
EXPERIMENT 1
The purpose ofExperiment 1 was twofold. First, it was
designed to test the assumptions ofsimple serial and parallel models, which do not predict multiple-feature benefits. Second, it was designed to determine whether multiple-feature benefits occur (1) when only a single object
is presented, (2) when the location of the object is constant, and (3) when the target feature or combination oftarget features varies across trials. The discrimination difficulty ofthe three different target features (color, shape,
and size) within a letter varied as described in the preliminary experiment. Speeded two-choice responses were
jects were instructed to respond as quickly and as accurately as possible. "Present"/"absent" response direction was held constant für
each subject.
Target features varied across trials. There were three possible
one-feature judgments (color, size, and shape), three possible combinations of two-feature judgments (shape and color, shape and
size, color and size), and one combination ofthree-feature judgments
(size, color, and shape). The eight possible letter probes resulting
from the color, size, and shape features were run within subjects,
and the specific target color, shape, and/or size to be judged (target
feature combinations) were run between subjects. The eight possible
target feature combinations were as folIows: large green S, small
green S, large green C, small green C, large red S, small red S, large
red C, and small red C. In order to minimize response competition
between trials, each subject was assigned to one ofthe above target
feature combinations. Thus, each subject was asked to make "present" judgments for only one specific target feature combination.
1386
FOURNIER, ERIKSEN, AND BOWD
The subjects received 16 practice trials in which the color, shape,
and size ofthe letter probe were verbally reported. Those who exceeded four errors in this task were excused from the experimental
sessions.
The subjects completed two experimental sessions consisting of
eight blocks of74 trials each. Each block consisted of 24 one-feature
judgrnents, 36 two-feature judgments, and 14 three-feature judgments. On half of each of the feature judgment trials, the probe contained the specific target features; on the other half of the trials, the
probe did not. The specific target features and the presence/absence
of the target features occurred randomly within a block.
Data analysis. Mean correct RTs and percent accuracy were analyzed. Repeated measures analysis of variance (ANOVAs) were
performed on the factors of response type (2 levels: "present" and
"absent") and feature judgment (7 levels: size, shape, color, size
and shape, size and color, color and shape, and three-feature). In
addition, repeated measures ANOVAs were conducted on correct
U
G)
U)
E
Results and Discussion
As shown in Figure 1, multiple-feature benefits occurred für "present" responses. Also, sornewhat weaker
multiple-feature benefits were found für "absent" responses. A repeated measures ANOVA (block [8] X ses-
Present Response
600
c:::::::J "Absent" Response
580
560
540
520
500
480
460
_
-
"absent" response trials for each ofthe two-feature judgments and
the three-feature judgments based on which target features were
present in the object [target feature(s) present]. The levels ofthe target feature(s) present factor were as folIows: the two-feature size
and shape judgment had three levels (shape, size, and none); the
two-feature size and color judgment had three levels (color, size,
and none); the two- feature color and shape judgment had three levels (color, shape, and none); the three-feature judgment had seven
levels (color and shape, size and color, size and shape, color, shape,
size, and none).
11
11
440
420
400
380
360
4
-
6
Ir'
8
...
e...
W
-e
c
Q.
1
!
10
12
14
I
G)
CI)
lf
16
18
20
.
Size Shape Color
Size
Size Color
Shape Color Shape
Three-feature
Feature Judgments
Figure 1. RTs and percent errors for "present" and "absent" responses for each of
the feature judgments in Experiment 1.
FEATURE DISCRIMINATION
sion [2] x response type [2] X feature judgment [7])
was conducted on both correct mean RTs and accuracy.
All significant effects for RTs and accuracy are presented in Table 1.
Practice effects. The main effects and interactions involving session and block for RT and accuracy indicated
that feature judgment performance improved with practice. In addition, the session X feature judgment interaction for RT indicated that the decrease in RT across
sessions was smaller for the color-alone judgment than
for the other feature judgments (25 msec vs. approximately 45 msec; Tukey, p < .05). This finding suggests
that the very fast RTs for the color-alone judgment was
limited by a floor effect in Session 2. Also, the session
X feature judgment interaction for accuracy indicated
that the errors for the size-alone judgment decreased
more (by 2%) relative to the shape and two-feature size
and color judgment (Tukey, ps < .05). More importantly,
multiple-feature benefits were found for "present" responses in the first block of trials in Session 1 (see
Table 2), even though the subjects received only 16 practice trials. This finding suggests that multiple-feature
benefits are not dependent on practice effects. Because
feature judgments for "present" and "absent" responses
did not vary across sessions, session effects are not discussed further.
Figure 1 shows that as the discriminability of the target features increased, RT and errors decreased (main effect of feature judgment). As predicted, "present" and
"absent" RTs among the one-feature judgments were
fastest and most accurate for color, slowest and least accurate for size, and intermediate in speed and accuracy
forshape [forRT, plannedF(l,7) > 9.0 < 20.0,p < .05;
for accuracy, planned F(l,7) > 4.20 < 19.90, P < .07].
In addition, "absent" responses were slower than "present' responses (main effect of response), and "absent'
responses became even slower and less accurate relative
Table 1
Correct Response Time and Accuracy ANOVA Tables
for Significant Effects in Experiment 1
Factor
Significance
MSe
response
feature
scssion
block
session X block
session X feature
block X feature
response X feature
Response Time
F(I,7) = 8.66,p < .05
F(6,42) = 26.62,p < .001
F(I,7) = 22.57,p < .001
F(7,49) = 9.82,p < .001
F(7,49) = 6.23, p < .001
F(6,42) = 3.47, p < .01
F(42,294) = 1.56,p < .05
F(6,42) = 5.96,p < .001
88,169
8,374
43,518
7,244
5,229
2,260
1,751
3,360
responsc
feature
block
session X feature
response X feature
session X block
Accuracy
F(I,7) = 6.57,p < .05
F(6,42) = 8.96,p < .001
F(7,49)=2.61,p<.05
F(6,42) = 2.71,p < .05
F(6,42) = 7.51,p = .001
F(7,49) = 2.70, p < .05
.0900
.0425
.0191
.0154
.0289
.0176
Note-feature = feature judgment.
1387
to "present" responses as the number of target features
increased (response X feature judgment). Because
trends across feature judgments varied between "present' and "absent" responses, these trends were evaluated within each response type by planned F comparisons, with 1 and 7 degrees offreedom (dfs).
"Present" responses. As predicted, multiple-feature
benefits were found for "present" responses and were dependent on the combination of target features (see Figure I). When the subjects were required to judge the
presence of a conjunction of features, in which one feature was relatively difficult to discriminate (e.g., size)
and one or more features were relatively easy to discriminate (e.g., color and/or shape), their RTs were faster
and more accurate than when they judged the presence of
the less discriminable feature alone (e.g., size). More
specifically, "present" responses for the two-feature size
and color and two-feature size and shape judgments were
faster (by 46 and 26 msec, respectively) and more accurate than the size-alone judgment (ps< .05). Also, the
two-feature color and shape judgment was faster (by
29 msec) and more accurate than the shape-alone judgment (for RT,p < .05; for accuracy,p = .056). Moreover,
the three-feature judgment was faster than the sizealone, shape-alone, and two-feature size and shape judgments (61,25, and 34 msec faster, respectively; ps < .05).
The three-feature judgment was also more accurate than
the size-alone and two-feature size and shape judgments
(ps < .01). There was no significant RT and accuracy
difference between the three-feature and two-feature size
and color judgments (ps> .08) or the two-feature color
and shape judgment (Fs < 1). As shown in Figure 1,
error data indicate that the RT effects were not due to a
speed-accuracy tradeoff.
"Absent" responses. Figure 1 shows that multiplefeature benefits for "absent" responses were not as large
as those found for "present" responses. Also, as previously mentioned, "absent" responses were substantially
slower and less accurate than "present" responses for the
two-feature and three-feature judgments, relative to the
one-feature judgments. This might have been due to decision competition (Proctor, 1981) or response competition (8. A. Eriksen & C. W. Eriksen, 1974); this is possible because sometimes one or more target features were
present in the object when an "absent" response was required in the multiple-feature judgments. For example,
"absent" responses were required in the three-feature
judgments when the letter probe contained one target
feature that was absent and two target features that were
present, when the letter probe contained two target features that were absent and one target feature that was present, and when all three target features were absent.
Perhaps the target feature(s) present within the object
activated the "present" response, which, in turn, interfered with activation ofthe "absent" response. This idea
is consistent with the concept of response competition
(e.g., see Coles, Gratton, Bashore, C. W. Eriksen, & Donchin, 1985; B. A. Eriksen & C. W. Eriksen, 1974; Miller,
1388
FOURNIER, ERIKSEN, AND BOWD
Table 2
Feature Judgments for "Present" Responses
in the First Block of Trials in Experiment 1
Feature Judgments
size & size & color &
size shape color shape color
shape
Reaction time (msec) 465 464
379
459
457
414
25
9
0
21
25
17
Errors (%)
1991). To evaluate this possibility, repeated measures
ANOVAs (letter probe X target feature[s] present) were
conducted on correct "absent" RTs and accuracy separately for the three-feature judgments and each of the
two-feature judgments. Figures 2 and 3 show "absent"
RTs and accuracy relative to which target features were
three-feature
418
11
present in the object for the three-feature and two-feature judgments, respectively.
These data show that when the most discriminable feature (i.e., color) was present, "absent" response performance suffered most in both the three-feature and the
two-feature judgments. There was a main effect oftarget
-
600
580
E 560
540
a:: 520
CI)
U)
500
e
0 480
Cl.
U) 460
CI)
a:: 440
eCI) 420
U) 400
.0 380
360
u
CI)
U)
-...
-
5
10
15
20
2
W
t:
25
u
30
Q)
Q)
a.
35
40
45
50
Color
Shape
Color
Size
Size
Shape
Color
Shape
Size
None
Target Features Present For The Three- Feature Judgment
Figure 2. Correct "absent" response RTs and percent errors according to which target features were present in the three-feature judgment condition in Experiment 1.
FEATURE DISCRIMINATION
_
600
580
560
I - 540
IX: 520
=
o
C
1389
Target Feature Present:
c:::::J Color
Shape
1:::3 Size
lZZZl None
l1Q2Ql
500
480
IX: 440
-C
G)
I/)
420
400
.a
380
=
360
c(
o
5
..
e.
-
10
15
W
C
20
G)
25
G)
Q.
30
35
40
Color & Shape
Color & Size
Size & Shape
Two- Feature Judgments
Figure 3. Correet "absent" response RTs and percent errors according to which target features
were present in each ofthe two-feature judgments in Experiment 1.
features present for both RT and accuracy for the threefeature judgment [for RT, F(6,42) = 14.04, and for accuracy, F(6,42) = 2.43, ps < .05], the two-feature shape
and size judgment [for RT, F(2,14) = 20.32, and for accuracy, F(2,14) = 30.85, ps < .001], the two-feature
color and size judgment [for RT, F(2,14) = 37.65, and for
accuracy, F(2,14) = 19.59,ps < .001], and the two-feature color and shape judgment [for RT, F(2, 14) = 13.39,
and for accuracy, F(2,14) = 9.55,ps < .01]. Significant
trends for the feature judgments were evaluated by
planned F comparisons, with land 7 dfs.
Significant trends for the three-feature judgments were
as folIows. When one of the three target features was
present, "absent" responses were slower when this feature was color relative to when it was shape or size (ps<
.01). When two ofthe three target features were present,
"absent" responses were slower when these features
were color and shape or color and size relative to when
they were size and shape (ps< .00 I). Errors tended to
decrease as RT decreased, indicating that the above RT
interpretations were not due to a speed -accuracy tradeoff.
Similar results were found for the two-feature judgments. "Absent" responses were slower (ps< .0 I) and
less accurate (ps< .064) for the two-feature size and
shape and two- feature color and size judgments when the
more discriminable target feature (i.e., color or shape)
was present and the less discriminable feature (i.e., size)
was absent. In addition, "absent" responses were slower
and less accurate for the two-feature color and shape
judgment when the more discriminable feature (i.e.,
color) was present and the less discriminable feature
(i.e., shape) was absent (ps< .05). Finally, "absent" responses for the two-feature judgments were overall
fastest (ps< .05) and most accurate (ps::::; .064) when
none of the target features were present.
1390
FOURNIER, ERIKSEN, AND BOWD
These results show that increasing the discriminability oftarget features present in the object reduced the speed
and accuracy ofthe correct "absent" response (multiplefeature costs). This finding suggests that decision or response competition often occurred in the multiplefeature judgments when an "absent" response was required.
Summary. Multiple-feature benefits were found for
"present" responses when an object was presented alone,
at a constant location, and when the target features
changed across trials. These results also showed that discrimination performance was dependent on the discriminability of target features that were present in the object. When the discriminability oftarget features present
in the object increased, correct "present" responses were
faster and more accurate, whereas "absent" responses
were slower and less accurate.'
Multiple-feature benefits are not consistent with simple serial and paral1el models. However, it is possible that
benefits are based on decision (e.g., Proctor, 1981; Ratcliff, 1978) or response priming (e.g., Miller, 1982a). This
is possible because the "absent" response data suggest
that decision or response competition played a role in
multiple-feature costs. Before these possibilities are considered, the robustness of multiple-feature benefits and
costs, as wel1as other possible explanations, wil1 first be
examined.
Recal1 that, in Experiment 1, the specific target features were run between subjects. In addition, the specific
letter probes requiring a "present" response occurred
more frequently as the number oftarget features increased.
For example, when one feature was a target, four of the
probes contained this feature, and each was presented one
eighth ofthe time. When two features were targets, the two
probes that contained these two features were presented
one fourth ofthe time. Finally, when three features were
targets, the probe that contained these three features was
presented half ofthe time. Thus, the letter probe matching the assigned three-feature target was presented twice
as often as any ofthe other letter probes. It has been shown
(see Nickerson, 1972, 1973) that increased frequency of
presentation of a particular stimulus (letter probe) can
facilitate RT to this stimulus.
EXPERIMENT 2
The purpose ofExperiment 2 was to determine whether
multiple-feature benefits are obtained when al1 possible
features and combination of features are discriminated
and when the frequency of each letter probe mapped to
the "present" response is equated across the one-, two-,
and three-feature judgments. In order to balance the frequency without markedly increasing the number of'trials,
the letter probes that did not contain any target features
were not presented in the multiple-feature judgments. As
a result, the total number of multiple-feature j udgment trials was reduced, relative to Experiment I. However, we
expected to find benefits ifthey were not linked to probe
presentation frequency differences across the feature
judgments.
Method
Subjects. One male and 7 female undergraduate and graduate
students from the University of Illinois served as paid volunteers.
All subjects reported having normal or corrected-to-norrnal vision.
Apparatus, Stimuli, and Procedure. The apparatus, stimuli,
and procedures were identical to those in Experiment I, with the
following exceptions. First, the message indicating which features
were targets was in a font different from that of the letter probe.
Second, each subject judged the presence or absence of each possible one-, two-, and three-feature combination. Third, the presentation frequency of each possible letter probe requiring a "present"
response was balanced across the one-, two-, and thrce-feature
judgments.
The eight possible letter probes were presented five times each
within each feature judgment in a session. Thus, each of the six onefeature judgments contained 40 letter probe presentations with a
"present":"absent" response ratio of 20:20, each of the 12 twofeature judgments contained 20 letter probe presentations with a
"present'Y'absent" response ratio of 10:10, and each of the eight
three-feature judgments contained 10 letter probe presentations
with a "present'Y'absent" response ratio of 5:5.
In order to equate "present" and "absent" responses across feature judgments, some letter probes were removed. In the two-feature
judgments, the two letter probes that did not contain either target
feature were not presented. For example, if the target features were
red S, then letter probes with a different color and shape were not
presented. In the three-feature judgments, two letter probes were
not presented: the letter probe that did not contain any of the three
target features, and the letter probe whose color and shape were different from the target color and shape.
Each experimental session consisted of one block of 560 trials
(80 trials of three-feature, 240 trials of two-feature, and 240 trials
of one-feature judgments). The subjects received 55 practice trials
in the first session. They took short breaks at l O-rnin intervals and
when fatigued.
Results and Discussion
Practice etTects. A repeated measures ANOVA (session [2] X response [2] X feature judgment [7]) was performed on mean correct RTs and accuracy. Al1 significant effects were similar to those in Experiment I. There
was a main effect ofsession for RTs [F(1,7) = 36.85,p <
.001] and accuracy [F(1,7) = 6.47,p < .05] and a session
X feature judgment interaction for RTs [F(6,42) = 3.0,
p < .05]. Responses were overal1 faster and more accurate in Session 2 than in Session I. In addition, RTs decreased more for the multiple-feature judgments (70 msec)
than for the one-feature judgments (40-50 msec) in Session 2. The above findings, along with the finding that
there was no significant difference between "present" and
"absent" responses across sessions, indicate that the 55
practice trials were not sufficient for the subj ects' to Iearn
al1 possible target-response mappings, especial1y when
discriminating multiple features. It might have taken much
longer to learn the response mappings in Experiment 2
than in Experiment I since responses were based on all
possible target feature combinations, which varied across
trials. For this reason, only Session 2 was analyzed.
"Present" and "absent" responses across the different
feature judgments are presented in Figure 4. Similar to
FEATURE DISCRIMINATION
CJ
CI)
fI)
E
I-
0:::
"Present" Response
600 _
c::==J "Absent" Response
580
560
540
520
500
480
460
440
420
400
380
360
0
'---
1391
'---
2
4
...
6
w
8
e...
C
CI)
CI)
11.
10
12
14
Size Shape Color
Size
Size Color
Shape Color Shape
Three-feature
Feature Judgments
Figure 4. RTs and percent errors for "present" and "absent" responses for each of
the feature judgments in Experiment 2.
Experiment 1, multiple-feature benefits were found for
"present" responses; however, these benefits were somewhat weaker. A repeated measures ANOVA (response
[2] X feature judgment [7] ) was performed on both mean
correct RTs and accuracy in Session 2. All significant
effects for RTs and accuracy were similar to those found
in Experiment 1, and these effects are presented in
Table 3.
As shown in Figure 4, "present" and "absent" response
RTs and accuracy among the one-feature judgments
were not significantly different between color and shape
(planned Fs < 1). This finding suggests that color and
shape did not differ in discriminability, which is inconsistent with the one-feature judgment trends found in
Experiment I. However, both RT and accuracy for coloralone and shape-alone judgments were faster and more
accurate than the size-alone judgments [planned F( 1,7) >
6.20,ps< .05, for both RT and accuracy]. This indicates
that similar to Experiment 1, size was the least discriminable feature. Because the trends across feature judgments varied between "present" and "absent" responses
(response X feature judgment), they were evaluated
within each response type by planned F comparisons,
with land 7 dft.
"Present" responses. As is evident in Figure 4, multiplefeature benefits were found for "present" responses. In
general, when the subjects judged the presence of a conjunction of features, in which one feature was relatively
difficult to discriminate and one feature was relatively
easy to discriminate, their RTs were often faster and/or
more accurate than when they judged the presence ofthe
less discriminable feature alone.
1392
FOURNIER, ERIKSEN, AND BOWD
Table3
Correct Response Time and Accuracy ANOVA Tables
for Significant Effects in Experiment 2
Factor
Significance
MSe
response
feature
response x feature
Response Time
F(l ,7) 5.24, p = .056
F(6,42) = I1.30,p < .001
F(6,42) = 4.71,p < .001
3,328
432
319
feature
response x feature
Accuracy
F(6,42) = 3.98,p < .01
F(6,42) = 2.78,p < .05
.0157
.0266
Note-feature = feature judgment.
Although these benefits were not as strong as those
found in Experiment 1, the results, including the insignificant data trends, are consistent with Experiment I.
More specifically, the two-feature size and color judgment was significantly faster (27 msec) than the sizealone judgment (p < .01). In addition, the three-feature
judgment was faster (34 msec) and more accurate than
the size-alone judgment (p = .06, and p < .0 I, respectively). However, the three-feature judgment was not significantly faster (only 6-7 msec faster) than the two-feature size and shape judgment and the two-feature size
and color judgment, but the three-feature judgment was
significantly more accurate (ps< .05). This finding
suggests that a benefit existed for the three-feature judgment, relative to these two-feature judgments. Also, the
two-feature size and shape judgment was not significantly faster (19 msec) than the size-alone judgment (p =
.09), but the trends were consistent with those in Experiment 1. Unlike in Experiment I, however, responses for
the three-feature and two-feature color and shape judgments were no different from those for the shape-alone
judgment (Fs < I).
Significant multiple-feature benefits were found only
for feature judgments in which size was a target feature.
The lack of significant differences found between the
color-alone and shape-alone judgments (F < I) might
have contributed to the lack ofmultiple-feature benefits
found for the three-feature and two-feature shape and
color judgments relative to the shape-alone judgment.
"Absent" responses. Similar to Experiment I, multiple-feature costs were found for "absent" responses. Figures 5 and 6 show the relationship between correct "absent' responses (RTs and accuracy) and the target
features present in the object for the three-feature and
two-feature judgments, respectively. Because there were
very few observations per cell in this analysis, "absent"
response data from both experimental sessions were used
(ns = 2 and 5 maximum observations per cell for the threefeature and two-featurejudgments, respectively).Repeated
measures ANOVAs(letter probe X target feature[s] present) were conducted on correct "absent" RTs and accuracy
separately for the three-feature judgment and each of the
two-feature judgments.
As shown in Figures 5 and 6, "absent" RTs and errors
decreased as the number and discriminability of target
features present increased. There was a main effect of
target features present for both RTs and accuracy in the
three-featurejudgment [F(4,28) = 12.81,p < .001, and
F(4,28) = 2.65, p = .054, respectively] and in the twofeature size and color judgment [F(1,7) = 32.14,p < .001,
and F(1,7) = 15.29, P < .01, respectively]. There was
also a main effect of target features present for RTs in
the two-feature size and shapejudgment [F(1,7) = 179.53,
P < .001]. In general, "absent" responses for these feature judgments were hindered when the most discriminable ofthe target features (i.e., color and/or shape) was
present in the object and the least discriminable of the
target features (i.e., size) was absent. There were no significant main effects found in the two-feature color and
shape judgment, indicating that these features did not
significantly differ in discriminability. There was also a
significant letter probe X target feature(s) present interaction for accuracy [F(7,49) = 2.25,p = .05]; however, no
interpretation is offered for this latter effect.
Summary. Similar to Experiment I, increasing the
number and discriminability oftarget features present in
the object led to multiple-feature benefits for "present"
responses and multiple-feature costs for "absent" responses. These results were found when all possible feature combinations were judged and the presentation frequency of a specific stimulus (letter probe) mapped to the
"present" response was equated across the different feature judgments. The magnitude of'benefits, however, was
less than that found in Experiment 1. Significant benefits were found only when size was a target feature. The
reduction in benefits (and costs) for the two-feature color
and shape judgment relative to the shape-alone judgment
might have been due to the insignificant difference in RTs
and accuracy between the color target and shape target
discriminations. Perhaps multiple-feature benefits do not
occur iffeatures are similar in discriminability. Note that,
although stimuli were the same as those used in Experiment 1, this does not indicate that discrimination was
constant between these two experiments because discriminability is subjective and will vary across observers.
One possible explanation for multiple- feature benefits
is that the subjects strategically weighted their decisions
on the more easily perceived target features (i.e., color
and shape, but especially color) relative to the more difficult target feature (i.e., size). Thus, the least discriminable target feature (i.e., size), although not completely
ignored during multiple-feature judgments, might not
have been evaluated on each trial. Thus, benefits may be
based on guesses leading to more false alarms for "present' responses.
This idea is consistent with the larger multiple-feature
benefits and costs found in Experiment 1 relative to Experiment 2. First, "present" response error rates were high
for the size-alone judgments and discrimination of the
size feature benefited most whenjudged in combination
FEATURE DISCRIMINATION
_
U)
600
-
560
CI)
U)
500
E
1393
580
t - 540
a:::
C
o
Q.
U)
CI)
520
480
460
a:::
440
CI)
420
400
=
1:
!c(
380
= 360 +--2 -,-----4
6
8
10
12
14
16
18
Color
Shape
Color
Size
Size
Shape
Color
Shape
Target Features Present
For Three-Feature Judgment
Figure 5. Correct "absent" response RTs and percent errors according to which
target features were present in the three-feature judgment in Experiment 2.
with color and/or shape in Experiment 1. Second, these
benefits were greater when "present" responses for the
size-alone judgment had a higher error rate (20% in Experiment 1) as opposed to a lower error rate (7% in Experiment 2). Third, "absent" response error rates for the
three-feature judgment was highest (44%) in Experiment 1 when only the target feature of size was mapped
to the "absent" response.
EXPERIMENT 3
The purpose ofExperiment 3 was threefold. The first
goal was to determine whether multiple-feature benefits
and costs are found when only the two features of color
and shape serve as targets.
The second goal was to determine whether the less
discriminable feature (i.e., shape) on a proportion oftrials is ignored or weighted less than the more discriminable feature (i.e., color) when subjects make a "present" judgment based on two features. We evaluated this
by adding two letters that flanked the target probe and
varying the response relevance of the flanker features.
These flankers contained features that were either response
compatible or response incompatible with the target
features (similar to the response-compatibility paradigm; B. A. Eriksen & C. W. Eriksen, 1974). Fm exampie, if the two-feature judgment was "red H" and the target probe was a red H, both flankers either were identical
to the target probe (e.g., red H) or contained a different
color (e.g., green H), a different shape (e.g., red K), or
1394
FOURNIER, ERIKSEN, AND BOWD
_
g
620
Target Feature Present:
600
E 580
; : 560
D:: 540
tn
Color
Shape
=
e
o
Q.
520
500
tn 480
CI)
D:: 460
_
440
CI)
420
e
!
400
380
-L-
_'_.....JJ=='_
____"....,.=_'
______1
O,---------,--nrxr-------r--r=,-------""""V'IF==,.---------,
...
5
e...
-...
w
eCI)
10
U
CI)
Q.
15
20
Color & Shape
Color & Size
Size & Shape
Two -Feature Judgments
Figure 6. Correct "absent" response RTs and percent errors according to which
target features were present in each of the two-feature judgments in Experiment 2.
a different color and shape (e.g., green K) than the target probe. If the shape feature was weighted less than
color for the two-feature color and shape judgment, we
should find less interference for "present" responses
when the flanker shape does not match the two-feature
judgment compared with when the flanker shape does
not match the one-feature judgment. In the former case,
shape may be less critical for the judgment (two thirds
probability of correct "present" response based on
color); in the latter case, the shape feature is critical for
the judgment.
The magnitude of flanker interference can be used as
a measure of feature weighting because attention can be
drawn to stimuli on the basis oftheir task relevance or attentional set (e.g., Cave & Wolf, 1990; Duncan & Humphreys, 1989; Egeth & Yantis, 1997; Treisman & Gelade,
1980). Thus, if the flanker contains task-relevant information, attention will be drawn to the flanker on the
basis ofthis information (see Johnston & Dark, 1986, for
a review). Cohen and Shoup (1997) have also shown that
target discrimination is delayed by a flanker ifthe flanker
contains a task-relevant feature that is incompatible with
the target feature of the to-be-attended target (probe).
Task-irrelevant flanker features were shown not to interfere with target discrimination of the probe. The Cohen
and Shoup findings suggest that attentional set can be
based on specific features within objects (see also negative priming based on incompatible flanker attributes;
Tipper, Weaver, & Houghton, 1994). Therefore, if attentional set is based on a single feature (i.e., color) on a proportion of trials, this attribute alone will influence selection on those trials. Also, if attentional set is biased
FEATURE DISCRIMINATION
tures are task relevant and on the basis of the discriminability of these relevant features.
for a particular feature (i.e., color), then the unbiased feature (i.e., shape) will have a lower selection weight and
should be less likely to influence selection.
The third goal was to determine whether flanker interference is influenced by the discriminability ofthe flanker
features that are incompatible with the target features
contained in the probe. If so, this suggests that features
of an irrelevant object are processed and interfere with
responses to the target object on the basis of which fea-
-
580
560
E
540
fI)
Method
Subjects. Nine male and 7 female undergraduates from Washington State University participated for optional research credit in
a psychology course. All reported having normal color vision, and
all had normal acuity as assessed by aSnellen chart.
Apparatus, Stimuli, and Procedure. The apparatus and stimuli were identical to those in Experiment I, with the following ex-
660 . - - - - - - - - - - - - - - - - - - - - - - - ,
A
640 Feature Judgment:
·0·· Shape
620
.•.. Color
600 .. • . Two-feature
CJ
CD
Q
...······Q··········2··········
2 !........ ..···1
2
520
500
480
460
!
!
!
440
420 +-__---.
None
Identical Shape diff Color diff Two-feature diff
Flanker Condition
660 . - - - - - - - - - - - - - - - - - - - - - - - ,
Feature Judgment:
B
640
D· Shape
..
..•.. Color
620
.•..
600
580
CJ
CD 560
fI)
E 540
520
i
;1.·······
....
.!
. ··t·
500
!
480
460
440
420
None
1395
Identical Shape diff Color diff Two-featurediff
Flanker Condition
Figure 7. RTs for "present" responses (panel A) and "absent" responses (panel B)
for each ofthe feature judgments across flanker conditions in Experiment 3. The levels
of flanker conditions were as folIows: none = no flanker; identical = flanker identical
to probe; shape diff = flanker shape different from probe; color diff = flanker color
different from probe; two-feature diff = both flanker features different from probe.
1396
FOURNIER, ERIKSEN, AND BOWD
Table4
Percent Errors for Each Feature Judgment Across
Flanker Condition for "Present" and
"Absent" Responses in Experiment 3
Flanker Condition
shape color two-feature
FeatureJudgment none identical
diff
diff
diff
shape
color
two-feature
"Present" Response
6.94
10.94
8.19
4.19
4.06
6.88
3.87
3.75
3.94
9.50
7.81
3.69
3.25
8.94
3.50
shape
color
two-feature
7.00
3.50
7.06
"Absent" Response
9.31
11.75
2.88
3.13
7.93
11.00
8.13
4.88
9.00
10.00
6.19
9.19
ceptions. The target probe could appear in any one of eight locations on an imaginary circle centered around the fixation cross. Two
white dashes (0.6° center-to-center cue-target distance) were used
as precues for target location and were presented 150 msec before
the appearance of the target. In addition, noise letters (flankers)
could appear simultaneously with the target probe at 1.15° ofvisual
angle to the left and right of the target probe. Target probe and
flankers appeared for 50 msec, 1.57° from the fixation cross. The
target probe and flankers were either Hs or Ks colored in red or
green. The flanker factor had five levels: (I) no flanker (none),
(2) flanker identical to target (identical), (3) flanker shape different
from target (shape diff), (4) flanker color different from target (color
diff), and (5) flanker color and shape different frorn target (twofeature diff). At the start of each block, a message displayed which
one or two target features to discriminate in the cued location.
The subjects completed a total offour sessions consisting of 12
blocks (4 blocks of each feature judgment). The first session was
practice and was not included in the data analyses. Feature judgments
(color, shape, color, and shape) were blocked, and block order was
counterbalanced across sessions within and between subjects. The
one-feature color and shape judgment blocks contained 40 trials,
and the two-feature judgment blocks contained 60 trials. Target
probe location and flanker condition occurred equally often in a
randorn order within blocks.
Results and Discussion
Figure 7 shows the RTs for each featurejudgment across
flanker condition for "present" responses (panel A) and
"absent" responses (panel B). Table4 contains the percent
errors for these factors. A repeated measures ANOVA(response [2] X feature judgment [3] X flanker [5]) was
conducted on mean correct RTs and accuracy.? All significant effects for RTs and accuracy were consistent
with those found in Experiments land 2 and are presented in Table 5.
As shown in Figure 7 and Table 4, multiple-feature
benefits were found for both "present" and "absent" responses (main effect of feature judgment). Specifically,
RTs were faster and more accurate for the color and shape
judgment (Ms = 534 msec and 6.3%, respectively) than
the shape-alone judgment (Ms = 569 msec and 9.1%, respectively) [for RT,plannedF(l,15) = 123.55,p < .001;
for accuracy, F(l,15) = 30.10,p < .001]. This finding
demonstrates that the multiple-feature benefits found in
Experiments land 2 were not dependent on the inclusion ofthe size feature judgment. In addition, it suggests
that multiple-feature benefits are not dependent on extremely high error rates for the less discriminable feature.
"Present" response error rates were 6%, 9%, and 4%,
and "absent" response error rates were 4%, 9%, and 9%,
for the color-alone, shape-alone, and two-feature judgments, respectively.
Also consistent with Experiments 1 and 2, the difference between "present" and "absent" responses was
greater for the two-feature judgment than for the shapealone and color-alone judgments (response X feature
judgment). This finding is consistent with the idea that
response or decision competition occurred on a proportion oftrials, inflating RTs and errors for the two-feature
"absent" responses. This finding is analyzed further in
the section on multiple-feature costs below.
Furthermore, the flanker features influenced target
feature RT and accuracy. As shown in Figure 7, the RT
effect of flanker was similar for "present" and "absent"
responses for the color-alone judgments, but the pattern
was quite different across response types for the shapealone and two-feature judgments (response X feature
judgment X flanker interaction). These patterns were
evaluated separately for "present" and "absent" responses
by planned F comparisons, with 1 and 15 dfe.
"Present" responses. As predicted, the flanker features
that were different from the target features contained in
the probe interfered with target feature discrimination
(see Figure 7, A). The color-alone judgment was 31 msec
slower when the flanker color was different from the target probe (color diff), relative to when no flanker was
present (none) (p < .01). Also, the shape-alone judgment
was 33 msec longer when the flanker shape was different from the target probe (shape diff), relative to when
no flanker was present (none) (p < .001). Similarly, the
two-featurejudgment was 36 msec longer when the flanker
shape was different from the target probe (shape diff),
relative to when no flanker was present (none) (p < .00I);
Table 5
Correct Response Time and Accuracy ANOVA Tables
for Significant Effects in Experiment 3
Factor
Significance
MS e
response
feature
flanker
response X feature
response X flanker
feature X flanker
response X feature
X
Response Time
F(I,15) = 63.90,p < .001 3,152
F(2,30) = 119.40, P < .00I 2,254
F(4,60) = 39.4I,p < .001
324
F(2,30) = 7.75,p < .01
684
F(4,60) = 3.85, p < .01
306
F(8,120) = 14.94,p < .001
242
flanker F(8,120) = 7.17,p < .001
217
Accuracy
F(2,30) = 34.34, p < .001
feature
F(4,60) = 3.42, p < .05
flanker
F(2,30) = 24.49, p < .00I
response X feature
F(8, 120) = 2.16, P < .05
feature X flanker
Note-feature = feature judgment.
.0020
.0017
.0020
.0019
FEATURE DISCRIMINATION
however, when the flanker color was different from the
target probe (color diff), RTs were not significantly increased (only 8 msec; p = .09). Finally, the two-feature
judgment was 24 msec longer when both color and shape
flanker features were different from the target probe (twofeaturediff), relative to when no flankerwas present (none)
(p < .001). The error data in Table 4 indicate that these
RT effects were not due to a speed -accuracy tradeoff.
As predicted, both the shape-alone and the twofeature judgments were similarly slowed (relative to the
no-flanker condition) when the flanker shape was different from the target probe. That is, when the flanker had
a different shape from the target probe, and shape was
one of the target features, RTs were slowed 36 msec for
the two-feature judgment and 33 msec for the shapealone judgment. This finding indicates that the shape
_
650
CI)
-
630
E
610
u
tII
discrimination was not weighted differently in these two
judgments. Also, error RTs showed no evidence that responses were based on fast guesses for the two-feature
judgment. Error RTs overall were slower than correct
RTs, and the pattern across feature judgments was similar to correct RTs. Mean error RTs were 501, 572, and
551 msec, for color, shape, and two-feature judgments,
respectively. Thus, multiple-feature benefits were not
dependent on decreased weighting of the less discriminable feature (i.e., shape) in the two-feature judgments.
"Absent" responses. As predicted, the flanker interfered with target feature discrimination within the probe
when the flanker contained the target features and the
probe did not. Figure 7B and Table 4 show that when the
flanker features differed from the probe, discrimination
performance decreased, relative to when no flanker (none)
Target Feature Present For Two-Feature Judgment:
c:::::J Color
lZS2Sl Shape
I - 590
570
a::CI)
tII 550
C
o
e,
:c
530
510
490
... 470
C
CI) 450
a::
.a
430
410
+---'---'--f"-"'-""'-.L-....,....l....L.JL-..L-----J...lf--I<--""---'------"-T'<LL....L......J--'*""<-...lL_----J
2
...
...e
w
C
CP
D.
4
6
8
10
12
14
16
18
20
22
24
26
28
30
+-__--...
None
1397
,-__--...
,-__-,-__---....J
Identical Shape diff Color diff Two-feature diff
Flanker Condition
Figure 8. Correct "absent" response RTs and percent errors across ßanker condition according to which target features were present in each of the two-feature judgments in Experiment 3.
1398
FOURNIER, ERIKSEN, AND BOWD
was present. First, the color-alone judgment was 30 msec
longer (no effect on accuracy) when the flanker color differed from the target probe (color diff) (p < .00 I). Second, the shape-alone judgment was 33 msec longer and
was less accurate when the flanker shape differed from
the target probe (shape diff) (ps< .0 I). Third, the twofeature judgment was 15 msec Ionger (no effect on accuracy) when the flanker color (color diff) or both the
flanker color and shape differed (two-feature diff) from
the target probe (ps< .01). Fourth, in the two-feature judgment, the flanker caused more errors (no effect on RT)
when its shape was different (shape diff) and its color
was the same as the target probe (p < .0 I). Although
RTs were not slower in the two-feature judgment when
the flanker shape was different (shape diff), the high
error rate is consistent with a speed -accuracy tradeoff interpretation. Thus, when the flanker had a different color
and/or different shape from the target probe, the flanker
interfered with the two-feature judgment. Taken together,
these results demonstrate that the response mapping of
flanker features along the relevant target dimensions influenced "absent" responses.
Multiple-feature costs. Consistent with a predictions,
Figure 8 shows that "absent" RTs and error rates for the
two-feature judgment increased as the number and discriminability oftarget features mapped to the "present" response increased. A two-way repeated measures
ANOVA (target features present [3] X flanker [5]) was
conducted on "absent" response RTs and accuracy for
the two-feature judgment. There was a main effect oftarget features present for RT [F(2,30) = 127.63,p < .001]
and accuracy [F(2,30) = 33.88,p < .001]. "Absent" RTs
were slowest and least accurate when the most discriminable feature (i.e., color) was present and the least discriminable feature (i.e., shape) was absent [for RT,planned
F(l,15) = 85.71; for accuracy, F(l,15) = 22.54; p».«:
.00 I]. "Absent" RTs and accuracy were intermediate
when the least discriminable feature (i.e., shape) was present and the most discriminable feature (i.e., color) was
absent [for RT, planned F(l,15) = 254.12; for accuracy,
F(l,15) = 45.91; ps < .001]. Finally, RTs were fastest
and most accurate when none ofthe features (i.e., color
and shape) were mapped to the "present" response [for
RT, planned F(l,15) = 24.67; for accuracy, F(l,15) =
30.l5;ps< .001].
In addition, RTs and/or errors increased when a target
probe feature(s) was mapped to the "absent" response and
the flanker feature(s) along the same dimension was
mapped to the "present" response. There was a main effect offlanker for RT [F(4,60) = 1O.69,p < .0001] and
accuracy [F(4,60) = 3.15,p < .05], and a target feature(s)
present X flanker interaction for RT [F(8,120) = 2.73,
p< .01] and accuracy [F(8,120) = 2.60, P = .01]. RTs
and errors increased as the discriminability ofthe target
features (mapped to the "present" response) contained in
the flankers increased [color diff vs. shape diff; planned
F(l, 15) = 12.40, P < .0 I]. As shown in Figure 8, this lat-
ter effect was influenced by which of the target probe
features were mapped to the "absent" response.
Summary. The results were consistent with the predictions. First, multiple-feature benefits and costs were
found for feature judgments based only on color and
shape. These findings indicate that the benefits and costs
found in Experiments land 2 were not dependent on the
inc1usion of a target feature that had low response accuracy (e.g., size). Second, there was no evidence indicating that the less discriminable feature (i.e., shape) was
weighted less than the more discriminable feature (i.e.,
color) in the two-feature judgments. When the flanker
shape was inconsistent with the target shape, "present"
response RT was equally inflated in the two-feature and
shape-alone judgments. Third, flanker interference was
based on whether a flanker feature(s) was task relevant
and whether this feature(s) was incompatible with a target feature(s) contained in the probe. This finding suggests that the irrelevant flanker features did not interfere
with target feature discrimination. Finally, "absent" responses for the two-feature judgments were influenced
by the discriminability ofthe target feature present in the
flanker. Correct "absent" RTs were longer when the more
discriminable target feature (i.e., color) was present in the
flanker, relative to the less discriminable target feature
(i.e., shape).
This latter finding suggests that the more discriminable
target feature contained in the flanker interfered with the
response decision earlier or began to prime its associated
response (i.e., "present" response) before the less discriminable target feature. Note, however, that the two-feature
"present" responses were not affected by the discriminability of the target features absent in the flanker (see
two-feature judgments in Figure 7A). These findings suggest that the subjects were more sensitive to flanker features that were consistent with the target feature judgment (i.e., flanker features that were consistent with a
"present" response, rather than an "absent" response). Furthermore, these findings again suggest that response or
decision priming may be responsible for multiple- feature
benefits and costs.
Another possibility, however, is that the multiplefeature benefits (and costs) found in Experiments I 3
might have been due to contingencies inherent in our experimental design. For example, one possible explanation for multiple-feature benefits is that the more discriminable target features in our stimulus sets were correlated
with the "present" response in the multiple-feature judgments. In Experiments the appearance ofthe relevant, most discriminable feature (e.g., the color red) predicted a "present" response more often than an "absent"
response in the arrays where multiple features were targets. For example, ifthe target features were large red S,
then the probe letters containing the most discriminable
feature (red) would predict a "present" response seven out
of nine times (Experiment I) and five out of eight times
(Experiment 2). Also, the target features red H in Ex-
FEATURE DISCRIMINATION
periment 3 had probe letters containing red four out of
six times. In contrast, the most discriminable feature
(red) would equally predict a "present" or "absent" response when the target feature was only large (Experiments land 2) or H (Experiment 3).
These probabilities are inherent in the stimuli sets used
when equating the probability of "present" and "absent"
responses. Multiple-feature benefits, however, cannot be
completely due to correlated feature biases, because we
found benefits to occur in the first block oftrials in Experiment I (see Table 2), even though the subjects received
only 16 practice trials beforehand.
Another possible explanation for multiple-feature benefits concerns the expected frequency of each target feature. The expected frequency of each target feature across
the one-, two-, and three- feature judgments was not held
constant in Experiments 1-3. For example, in Experiment 1, the expected frequency of the target feature on a
one-feature judgment trial was .50, because half of the
trials contained a target feature. On a two-feature trial,
the expected frequency of each target feature was above
.50, because a target feature appeared in all but two target-absent displays. On a three-feature trial, the expected
frequency of each target feature was weil above .50, because all but one ofthe target-absent displays contained
at least one target feature. The expected frequency of
each target feature also increased for the multiplefeature judgments in Experiments 2 and 3. Perhaps
multiple-feature benefits are based on two things: the
discriminability of the target feature, and the expected
frequency of a target feature. If identification time decreases as the expected frequency of the target feature
increases, multiple-feature benefits may be due to this
contingency'
EXPERIMENT 4
To ensure that the multiple-feature benefits found in
Experiments 1-3 were not due to correlated feature biases or to the expected target frequency contingencies
described above, Experiment 4 was conducted. Experiment 4 was similar to Experiment 3, with the exception
that each possible letter stimulus (red H, green H, red K,
green K) appeared equally often across each ofthe onefeature and two-feature judgments. Thus, for the twofeature judgments, stimuli requiring a "present" response
appeared only one fourth ofthe time, and stimuli requiring an "absent" response appeared three fourths of the
time. It was predicted that if correlated feature biases or
expected target frequency contingencies were responsible for the multiple-feature benefits found previously,
these benefits should disappear when these biases and
contingencies are eliminated.
Method
Subjects. Nine male and 7 female undergraduates from Washington State University participated for optional research credit in
a psychology course. All reported having normal color vision, and
all had normal acuity assessed by aSnellen chart.
1399
Apparatus, Stimuli, and Procedure. The experiment was identical to Experiment 3, with the following exceptions. Each probe
stimulus appeared equally often across each ofthe one-feature and
two-feature judgments. Thus, the color feature and the shape feature predicted a "present" response half ofthe time. However, three
fourths of the two-feature judgment trials required an "absent" response, and only one fourth of these trials required a "present" response. The subjects completed 12 blocks of 40 trials in each of
four sessions. Session 1 was practice and was not included in the
data analyses.
Results and Discussion
In general, the results were similar to those found in
Experiment 3 (see Figure 9 and Table 6 for RTs and errors,
respectively). A repeated measures ANOVA (session [3]
X response [2] X feature judgment [3] X Flanker [5])
was conducted on mean correct RTs and accuracy.' The
significant main effects and interactions are presented in
Table 7.
Only results related to multiple-feature benefits are discussed. Figure 10 shows the RT and errors for "present"
and "absent" responses for each ofthe feature judgments,
collapsed over flanker condition.
As shown in Figure 10, "present" and "absent" response RTs and accuracy varied across the different feature judgments. "Present" responses were fastest for the
color-alone judgment [planned F(l, 15) = 60.54, P <
.001]. In addition, unlike in Experiment 3, there was no
significant RT difference between the two-feature and
shape-alone judgments for "present" responses (p =
.16); however, the two-feature judgment was more accurate (by 4%) than the shape-alone judgment [planned
F( 1,15) = 17.10, P < .001]. "Present" responses for the
two-feature judgment were as accurate as those for the
color-alone judgment (planned F < 1). That is, although
there was no difference in "present" RTs for the shapealone and two-feature judgments, error rates were significantly higher for the shape-alone judgment. Thus, it
appears that the RT for the shape judgment was lowered
bya speed-accuracy tradeoff, given the high error rate.
Further support for this speed -accuracy tradeoff was
found when examining error RTs. Mean error RTs were
438,492, and 486 msec, for color, shape, and two-feature
judgments, respectively. Error RTs were approximately
20 msec faster than correct RTs; however, error RTs and
correct RTs showed the same pattern across feature judgments. Faster RTs for errors indicate that errors were
mainly due to fast guesses, and it appears that these fast
guesses occurred significantly more often for the shapealone judgment.
These findings indicate that multiple-feature benefits
based on accuracy were found for the two-feature judgment, relative to the shape-alone judgment. Also, a possible speed-accuracy tradeoff for the shape judgment
might have been responsible for the reduced benefits
found for RT.
The "absent" response data were similar to those found
in Experiment 3. "Absent" responses were fastest and most
accurate for the color-alone judgment [for RT, planned
F(l ,15) = 70.81,p < .001; for accuracy, F(l, 15) = 12.57,
1400
FOURNIER, ERIKSEN, AND BOWD
660 , . . - - - - - - - - - - - - - - - - - - - - - - ,
A
640 Feature Judgment:
o Shape
620
•
Color
600
580
CJ
CI) 560
tn
E 540
520
500
480
460
440
-
..•.. Two-feature
420 + - - - - - - - r - - - - - , - - - - , - - - - - - , - - - , - - - - '
None
Identical Shape diff Color diff Two-featurediff
Flanker Condition
-
660 , - - - - - - - - - - - - - - - - - - - - - - ,
B
640 Feature Judgment:
··0·· Shape
620
..•.. Color
..•.. Two-feature
600
580
CJ
560
tn 540
E
CI)
.
.'
......
520
500
480
460
440
420 +-__-,
None
···t
--=;-
-y--_ _
,..--_ _-t--r-_ _--...-J
Identical Shape diff Color diff Two-featurediff
Flanker Condition
Figure 9. RTs for "present" responses (panel A) and "absent" responses (panel B)
for each of the feature judgments ac ross f1anker conditions in Experiment 4. The levels
of f1anker conditions were as folIows: none = no flanker; identical = f1anker identical
to probe; shape diff = f1anker shape different from probe; color diff = f1anker color
different from probe; two-feature diff = both f1anker features different from probe.
p< .01], intermediate for the two-feature judgment, and
slowest and least accurate for the shape-alone judgment
[for RT,planned F( 1,15) = 66.94, P < .001; for accuracy,
F(1,15) = 5.31,p < .05].
However, unlike Experiment 3, "absent" responses
were faster than "present" responses for the two-feature
judgment (response X feature judgment). This was most
likely due to the higher probability of "absent" responses
(75%), relative to "present" responses (25%), for the twofeature judgment. Perhaps the criterion for an "absent"
response for the two-feature judgment was lower than
the criterion for a "present" response.
In sum, multiple-feature benefits (in terms ofaccuracy)
were found for "present" responses when each possible
letter stimulus (red H, green H, red K, green K) appeared
equally often across each of the one-feature and twofeature judgments. Thus, multiple- feature benefits found
in Experiments 1-3 were not completely due to correlated feature biases or contingencies based on expected
target frequency. In addition, benefits were found for the
FEATURE DISCRIMINATION
Table6
Percent Errors for Each Feature Judgment
Across Flanker Condition for "Present" and
"Absent" Responses in Experiment 4
Flanker Condition
shape color two-feature
diff
diff
Feature Judgment
none identical
diff
shape
color
two-feature
"Present" Response
7.21
6.69
13.38
4.31
4.65
5.48
4.81
7.75
5.08
9.09
10.00
6.35
14.09
8.64
8.08
shape
color
two-feature
5.67
2.52
5.00
"Absent" Response
3.64
8.48
1.52
1.75
3.69
4.54
3.60
4.44
3.37
5.71
4.37
4.58
Table 7
Correct Response Time and Accuracy ANOVA Tables
for Signlflcant Effects in Experiment 4
Factor
Significance
MSe
Response Time
F(2,30) = 9.12, P < .001
6,572
session
feature
F(2,30) = 92.97,p < .001 5,798
flanker
F(4,60) = 62.22, P < .00 I
795
response X feature
F(2,30) = 23.73, p < .001 2,026
feature X flanker
F(8,120) = 17.66,p < .001
626
response X feature X flanker
F(8,120) = 3.15,p < .01
645
response
feature
flanker
response X session
response X flanker
session X flanker
feature X flanker
Accuracy
F(l,15) = 14.27,p < .01
F(2,30) = 23.72,p < .001
F(2,30) = 10.20,p < .001
F(2,30) = 3.69, p < .05
F(4,60) = 6.37,p < .001
F(8, 120) = 2.00, P = .05
F(8,120) = 4.04,p < .001
.0286
.0064
.0047
.0037
.0044
.0039
.0056
Note-feature = feature judgment.
two-feature judgment even though "present" responses
were less correlated with the correct response (only 25%
ofthe trials). There was, however, a reduction in RT benefits, relative to what was found in Experiments 1-3.
Thus, it is possible that correlated feature biases and contingencies based on expected frequencies contribute to
multiple-feature benefits.
GENERAL DISCUSSION
The results of these experiments demonstrate that
judging the presence of multiple features, in which one
feature is relatively difficult to discriminate and one or
more features are relatively easy to discriminate, can be
faster and more accurate thanjudging the presence ofthe
less discriminable feature alone (multiple-feature benefits). In addition, judging the absence of one or more
multiple target features led to slower and less accurate
"absent" responses as the number and discriminability of
target features present in the object increased (multiplefeature costs). Multiple-feature benefits and costs were
found when several different target feature combinations
1401
were discriminated or a subset oftarget feature combinations was discriminated. In addition, benefits and costs
were found when the target object appeared alone or
among irrelevant objects and when the target appeared in
a fixed or random spatiallocation.
Factors Not Responsible for Benefits and Costs
Several explanations of multiple- feature benefits and
costs were ruled out in this study. First, Experiment 2
showed that benefits and costs were not due to differential presentation frequency ofthe target features or combination oftarget features. Benefits and costs were found
when each subject judged the presence and absence of all
possible target feature combinations across trials. However, benefits were not as strong as those found in Experiments 1 and 3. Perhaps this was because target features were consistently mapped in the multiple-feature
judgments in Experiments 1 and 3 but not in Experiment 2. This finding suggests that reducing the featureto-response mapping ratio may lead to more robust
multiple-feature benefits.
Second, Experiment 3 showed that benefits and costs
were not due to the subjects' strategically weighting their
decisions on the more discriminable target features. The
least discriminable feature (i.e., shape) was weighted
equally when judging the presence of this feature alone
or in conjunction with a more discriminable feature (i.e.,
color). Specifically, "present" response RTs were equally
inflated in the two-feature color and shape judgments
and in the shape-alone judgments when the flanker's least
discriminable feature (i.e., shape) was inconsistent with
the correct "present" response. Error RTs also indicated
that "present" responses were not based on fast guesses
in the two-feature judgments. Thus, it is unlikely that benefits were based on guesses leading to more false alarms.
Third, Experiment 4 showed that multiple-feature
benefits were not completely due to correlated feature
biases or to contingencies based on differential target feature expectancies across the feature judgments. When
each feature within the target object equally predicted a
"present" or "absent" response in single- and multiplefeature judgments, multiple-feature benefits were observed. However, benefits were based on accuracy only.
The lack of multiple-feature benefits based on RT might
have been a result of the fast RTs and high error rates
found for the less discriminable target feature (i.e., shape
judgment), which was likely due to a speed-accuracy
tradeoff.
In addition, the reduction in RT benefits in Experiment 4 might have been due to a decrease in the correlation between the "present" response and the correct response in the multiple-feature judgments. "Present"
responses occurred on only 25% ofthe multiple-feature
judgment trials, whereas they occurred on 50% of the
single-feature judgment trials. This reduction in "present" response frequency for multiple-feature judgments
might have led to an "absent" response bias, resulting in
slower "present" responses. While this explanation and
1402
FOURNIER, ERIKSEN, AND BOWD
CJ
G)
U)
E
600 Feature Judgment:
r::::::::::J Shape
580
Color
_
560
tw»""j Two-feature
540
520
500
480
460
440
420
400
380
360 +---------'---'2
.
g
W
cG)
,-----...---r
4
6
8
10
G)
D.
12
14
16 + - - - - - - - - - - r - - - - - - - - - - - r - - - - - - - - - '
"Present"
"Absent"
'Response
Figure 10. RTs and percent errors for "present" and "absent" responses for each of
the feature judgments collapsed over ßanker conditions in Experiment 4.
a possible speed-accuracy interpretation might account
for reduced benefits based on RT,a possible contingency
explanation cannot be completely ruled out. This is because multiple-feature benefits were larger in Experiments 1 and 3, in which these biases and contingencies
were the greatest.
Simple Serial and Parallel Models
The results ofthe present study bear on the validity of
simple serial and parallel models offeatureprocessing: That
is, such models have difficulty accounting for multiplefeature benefits. However, a simple parallel model may
be consistent with multiple-feature benefits if benefits
are based on decision or response priming by each target
feature, and this priming is combined to reach adecision
or response criterion (e.g., Miller, 1982a, 1982b). This
model is referred to as the asynchronous priming (AP)
model. In addition, the number of possible stimuli mapped
to the "present" response (e.g., Nickerson, 1973) or the
number of memory set items compared against the target object mayaiso contribute to multiple-feature benefits (e.g., Ratcliff, 1978). Before discussing these possibilities, the flanker results in Experiment 3, which provide
insights into multiple-feature benefits, will be discussed.
Flanker Interference and Insights
Into Benefits and Costs
Flanker interference in Experiment 3 was based on
whether a flanker feature(s) was task relevant and whether
the feature was incompatible with a target feature contained in the probe. For example, when shape was the target feature and only the flanker color was different from
the probe, the flanker color did not interfere with the
probe shape discrimination. This finding suggests that
FEATURE DlSCRIMINATION
the irrelevant flanker features did not interfere (or interfered very little) with target feature discrimination
within the probe. This is consistent with past research
that shows that irrelevant items are not analyzed in as
much detail as relevant items (e.g., Rabbit, 1967).
In addition, when the flanker features matched the target features but were incompatible with the correct "absent' response, the amount of interference depended on
target feature discriminability. For example, when the
target color was present in the flanker, it interfered more
with "absent" responses, relative to when the target shape
was present in the flanker. This finding suggests that interference occurred earlier when the target feature contained in the flanker was more discriminable. It is possible that the more discriminable target feature primed its
associated response earlier than the less discriminable
target feature. Discrimination RT benefits have been
shown to increase as the priming lead time increases for
the correct response (Flowers & WiIcox, 1982). Also, this
assumption is consistent with the findings ofGrice, Boroughs, and Canham (1984), who showed that discrimination of target stimuli affects the rate of growth of associative strength for the correct response.
Furthermore, evidence consistent with response priming is found when evaluating the multiple-feature costs
for "absent" responses in Experiments 1-4. Multiplefeature costs were based on the discriminability of the
target features that were mapped to the "present" response.
Specifically, increasing the discriminability of target
features mapped to the "present" response led to greater
increases in RTs and errors for "absent" responses. For
example, when the target features color and shape were
mapped to the "present" response, "absent" responses
were longer and less accurate, relative to when only the target feature shape was mapped to the "present" response.>
Possible Models That Can Account for
Multiple-Feature Benefits
An AP model (similar to the asynchronous discrete
model discussed by Miller, 1982a),6 a response mapping
model (see, e.g., Hick, 1952; Hyman, 1953), and a memory comparison (template) model (see, e.g., Ratcliff,
1978) may account for multiple-feature benefits. The AP
model assumes that all features within an object are processed in a parallel, independent fashion, and each target
feature primes its task-relevant response (e.g., "present"
or "absent"). Target features that are relatively easy to
discriminate (e.g., color) are assumed to prime their responses on average before features that are more difficult to discriminate (e.g., size; see also Grice et aI., 1984;
Grice, Canham, & Schafer, 1982). Furthermore, it is assumed that priming from the different features are combined to meet the response criterion (Fournier & C. W.
Eriksen, 1990; Miller, 1982a, 1982b). This may lead to
multiple-feature benefits when all of the target features
are mapped to the same response and may lead to multiplefeature costs when target features are not mapped to the
1403
same response. Thus, the speed and accuracy ofresponses
depend on the discriminability of the target features
mapped to the same or different response and mayaIso
depend on the strength of the target feature-response
mapping.
In contrast, the response mapping model predicts
multiple-feature benefits based on the number ofpossible stimuli mapped to the "present" response. In our study,
the number ofpossible stimuli (letter probes) that matched
the target feature description decreased as the number of
target features increased. For example, when the features
large red S were targets, only one stimulus alternative
(large red S) matched this description. When the feature
large was a target, four different stimulus alternatives
(large red S, large green S, large red C, or large green C)
matched this description. Decreasing the number ofpossible stimulus alternatives mapped to a single response is
known to decrease response latency and errors (see Nickerson, 1973, for a review). This factor, along with the
assumption that more discriminable target features prime
responses earlier than the less discriminable features
(see Kahneman, 1973), may account for multiple-feature
benefits.
Memory comparison, or template, models (see Farell,
1984; Ratcliff, 1978; Sternberg, 1969; see also Prinz &
Scheerer-Neumann, 1974) also assume that RTs are influenced by the set of stimulus alternatives mapped to
the positive response (e.g., "present" response). The positive stimulus alternatives are each represented by a template in memory, and a stimulus must be compared with
these templates before a positive or negative response
can be executed. Thus, reducing the number of stimulus
alternatives (templates) mapped to the positive response
will reduce response RTs. A template model, such as the
memory retrieval model proposed by Ratcliff (1978),
can account for multiple-feature benefits and costs.
This model assurnes that probe and memory set item
(template) features are compared in a parallel, independent, and continuous (random-walk) fashion. In addition, probe-template comparison time is dependent on
(I) the discriminability among the memory set items and
the distractors, (2) the order of the feature comparisons
and the distribution of matehing and nonmatehing features within that order (Ratcliff, 1978, p. 63), and (3) the
decision boundaries for "present" and "absent" responses
(which vary according to the information required for a
decision). Thus, multiple-feature benefits and costs may
be accounted for by assuming that the order of probetemplate feature comparison is biased toward comparing
the most discriminable target features with the template
first, followed by the less discriminable target features,
and then followed by any nontarget features (order of
comparison which mayaIso be based on discriminability).
The important assumption is that object features are
compared asynchronously on the basis of feature discriminability, and evidence indicating a match or mismatch
is combined to meet adecision criterion. Multiple-feature
1404
FOURNIER, ERIKSEN, AND BOWD
benefits will occur if all ofthe target features are present,
and costs will occur when some target features are present and some are absent.
These three models can also account for the weaker
benefits found in Experiments 2 and 4 by assuming response priming is reduced or the "present" response criterion is increased (e.g., Nickerson, 1973; Rabbit, 1967)
when increasing the number of targets mapped to a specific response (Experiment 2), when increasing the possible responses linked to a single stimulus (Experiment 2),
or when reducing the probability of a particular response
(Experiment 4). The AP model predicts that these weaker
benefits are based on a reduction in response priming. In
contrast, the memory retrieval model (e.g., Ratc1iff, 1978)
predicts that these weaker benefits are based on a higher
"present" response criterion. Furthermore, the response
mapping model can account for weaker benefits based
on reductions in response priming and/or an increase in
the "present" response criterion. Note, however, that the
AP model, but not the response mapping model or the
memory retrieval model, has difficulty accounting for
benefits based on accuracy alone in Experiment 4. The
AP model could account for the findings in Experiment 4
if the lack of RT benefits are due to a speed -accuracy
tradeoff.
Conclusions
In sum, our results imply that the final output of processing (response) can be faster when judging the presence ofmultiple features, as opposed to a single feature.
These findings are not consistent with simple serial and
parallel models of feature processing. However, models
that assume that responses or decisions are asynchronously primed by each target feature (based on feature
discriminability) and are combined to meet a response or
decision criterion may account for multiple-feature benefits (e.g., the AP and memory retrieval models). Also,
models that assume that both the number of stimuli
mapped to a particular response and the discriminability
of target features influence response or decision times
can account for multiple-feature benefits. Finally, other
explanations based on contingencies (e.g., expected target frequencies) mayaiso contribute to multiple-feature
benefits, since benefits were stronger when these contingeneies were present. Achallenge for future research
will be to determine the robustness of multiple-feature
benefits and the processes underlying this phenomenon.
REFERENCES
BAMBER, D. (1969). Reaction times and error rates for "sarne'<-vdifferent" judgments of multidimensional stimuli. Perception & Psychophysics, 6,169-174.
CAVE, K. R., & WOLFE, J. M. (1990). Modeling the role ofparallel processing in visual search. Cognitive Psychology, 22, 225-271.
COHEN,A., & SHOUP, R. (1997). Perceptual dimensional constraints in
response selection processes. Cognitive Psychology, 32, 128-181.
COLES, M. G. H.. GRATTON, G., BASHORE, T. R., ERIKSEN, C. W., &
DONCHIN, E. (1985). A psychophysiological investigation ofthe continuous flow model of human information processing. Journal of
Experimental Psychology: Human Perception & Performance, 11,
529-553.
COLES, M. G. H., SCHEFFERS, M. K., & FOURNIER, L. (1995). Where did
you go wrong? Errors, partial errors, and the nature ofhuman information processing. Acta Psychologica, 90, 129-144.
DUNCAN,J., & HUMPHREYS, G. (1989). Visual search and stimulus similarity. Psychological Review, 96, 433-458.
EGETH, H. E., & YANTIS, S. (1997). Visual attention: Control, representation, and time course. Annual Review ofPsychology, 48, 269-297.
ERIKSEN, B. A., & ERIKSEN, C. W. (1974). Effects ofnoise letters upon
the identification of a target letter in a nonsearch task. Perception &
Psychophysics, 16,143-149.
ERIKSEN, C. W., & SCHULTZ, D. W. (1979). Information processing in
visual search: A continuous flow conception and experimental results. Perception & Psychophysics, 25, 249-263.
FARELL, B. (1984). Attention in the processing of complex visual displays: Detecting features and their combinations. Journal ofExperimental Psychology: Human Perception & Performance, 10,40-64.
FLOWERS, J. H., & WILCOX, N. (1982). The effect offlanking context on
visual classification: The joint contribution of interactions at different processing levels. Perception & Psychophysics, 32, 581-591.
FOURNIER, L. R., & ERIKSEN, C. W. (1990). Coactivation in the perception ofredundant targets. Journal ofExperimental Psychology: Human
Perception & Performance, 16,538-550.
FOURNIER, L. R., &ERIKSEN, C. W. (1991, November). All-or-none versus a graded process conception ofattention. Poster presented at the
annual meeting of the Psychonomic Society, San Francisco.
GRATTON, G., COLES, M. G. H., SIREVAAG, E. J., ERIKSEN, C. W.. &
DONCHIN, E. (1988). Pre- and post stimulus activation of response
channels: A psychophysiological analysis. Journal of Experimental
Psychology: Human Perception & Performance, 14,331-344.
GRICE, G. R., BOROUGHS, J. M., & CANHAM, L. (1984). Temporal dynamics ofassociative interference and facilitation produced by visual
context. Perception & Psychophysics, 36, 499-507.
GRICE, G. R., CANHAM, L., & SCHAFER, C. (1982). Development of associative and perceptual interference. Perception & Psychophysics,
32,375-387.
HAWKINS, H. L. (1969). Parallel processing in complex visual discrirnination. Perception & Psychophysics, 5, 56-64.
HICK, W. E. (1952). On the rate of gain ofinformation. Quarterlv Journal ofExperimental Psychology, 4, 11-26.
HYMAN, R. (1953). Stimulus inforrnation as a determinant ofreaction
time. Journal ofExperimental Psychology, 45, 188-196.
JOHNSTON, W. A., & DARK, V. J. (1986). Selective attention. Annual Review ofPsychology, 37, 43-75.
KAHNEMAN, D. (1973). Attention and effort. Englewood Cliffs, NJ:
Prentice-Hall.
LOCKHEAD, G. R. (1972). Processing dimensional stimuli: A note. Psvchological Review, 79, 410-419.
LOCKHEAD, G. R. (1979). Holistic versus analytic process models: A
reply. Journal of Experimental Psychology: Human Perception &
Performance, 5, 746-755.
MILLER, J. O. (1978). Multidimensional same-differentjudgments: Evidence against independent comparison of dimensions. Journal of
Experimental Psychology: Human Perception & Performance, 4,
411-422.
MILLER, J. O. (l982a). Discrete versus continuous stage models of
human information processing: In search ofpartial output. Journal of
Experimental Psychology: Human Perception & Performance. 8,
273-296.
MILLER, J. O. (I 982b). Divided attention: Evidence for coactivation
with redundant signals. Cognitive Psychology, 14,247-279.
MILLER, J. [0.] (1991). The flanker compatibility effect as a function of
visual angle, attentional focus, visual transients, and perceptualload:
A search for boundary conditions. Perception & Psychophvsics, 49,
270-288.
MILLER, J. 0., & BAUER, D. W. (1981). Irrelevant differences in the
FEATURE DISCRIMINATION
"same't-vdifferent" task. Journal ofExperimental Psychology: Human
Perception & Performance, 7,196-207.
NICKERSON, R. S. (1972). Binary-classification reaction time: Review
of some studies of human information-processing capabilities. Psychonomic Monograph Supplements, 4, 275-317.
NICKERSON, R. S. (1973). Frequency, recency, and repetition effects on
same and different response times. Journal of Experimental Psychology, 101, 330-336.
OSMAN, A., BASHORE, T. R., COLES, M. G. H., & DONCHIN, E. (1992).
On the transmission ofpartial information: 1nferences from movementrelated brain potentials. Journal ofExperimental Psychology: Human
Perception & Performance, 18,217-232.
PRINZ,W., & SCHEERER-NEUMANN, G. (1974). Component processes
in multiattribute stimulus classification. Psychological Research, 37,
25-50.
PROCTOR, R. W. (1981). A unified theory for matching-task phenomena. Psychological Review, 88, 291-326.
RABBIT, P. M. A. (1967). Learning to ignore irrelevant information.
American Journal ofPsychology, 80, 1-13.
RATCLlFF, R. (1978). A theory ofmemory retrieval. Psychological Review, 85, 59-108.
SANDERS, A. F. (1981). Stage analysis ofreaction processes. In G. E.
Stelmach & 1. Requin (Eds.), Tutorials in motor behavior: Advances
in psychology (Vol. I, pp. 331-354). Amsterdam: North-Holland.
SMID, H. G. O. M., MULDER, G., & MULDER, L. J. M. (1990). Response
competition and the continuous flow conception ofhuman information processing: A psychophysiological investigation. Acta Psychologica, 74, 169-201.
STERNBERG, S. (1969). The discovery of processing stages: Extension
ofDonder's method. Acta Psychologica, 30, 276-315.
lipPER, S. P., WEAVER, 8., & HOUGHTON, G. (1994). Behavioural goals
determine inhibitory mechanisms of selective attention. Quarterly
Journal ofExperimental Psychology, 47, 809-840.
TREISMAN, A. M., & GELADE, G. (1980). Feature integration theory of
attention. Cognitive Psychology, 12, 97-136.
WICKENS, C. D. (1984). Engineering psychology and human performance. Columbus, OH: Merrill.
1405
NOTES
I. Multiple-feature benefits and costs were also found when stimulus sets (instead of messages) were presented, and the subjects determined whether the probe stimulus was a member of the stimulus set
(similar to aSternberg paradigm; e.g., Sternberg, 1969).
2. When the factor of session was included, all effects were identical
to those found for the three-way ANOVA. However, the main effect of
session approached significance [F(2,30) = 2.77, p = .078], and there
was a significant session x response X flanker interaction for RTs.
Flankers that had a different shape from the target showed decreased
RT interference across sessions for "absent" responses, but they were no
different across sessions for "present" responses. In addition, there was
a session x response X feature judgment interaction for accuracy
[F(4,48) = 2.70,p < .05], indicating that "present" responses for shape
judgments were less accurate for Session 4, whereas "absent" response
accuracy for shape judgments did not change across sessions.
3. Toby Mordkoff, personal communication, December 1996.
4. The main effect of session for correct RTs and a response X session interaction for accuracy indicate that RTs were faster and "present"
response accuracy was better during the last session.
5. The only exception to this example was found in Experiment 2,
which showed no difference in "present" response RT for the coloralone and shape-alone judgments. Also, there was no significant difference in multiple-feature costs for "absent" responses when color
alone or shape alone was mapped to the "present" response. These findings suggests that these two features were not different in discriminability. However, multiple-feature benefits and costs based on discriminabiiity were found in Experiment 2 (see Figures 4, 5, and 6).
6. Whether features are processed discretely and activate their associated responses (e.g., Miller, 1982a; Sanders, 198 I) or are processed
in a continuous fashion and activate their associated responses before
feature evaluation is complete (e.g., Coles et al., 1985; Coles, Scheffers, & Fournier, 1995; C. W. Eriksen & Schultz, 1979; Gratton, Coles,
Sirevaag, C. W. Eriksen, & Donchin, 1988; Grice et al., 1984; Osrnan,
Bashore, Coles, & Donchin, 1992; Smid, Mulder, & Mulder, 1990) will
not be debated.
(Manuscript received February 15, 1996;
revision accepted for publication November 8, 1997.)